Does a float need to be converted to network order?

[sidewinder]

If it does, and it is going between a little endian and a big endian
machines can we byte swap it as an int or long? Or does the mantissa or
exponent cross 2 byte boundarys?

Can I just call htonl ( (long) floatvalue))

 

[Adie]

The right question would be: 'does a float need to be converted to network
order if one want to <insert your action here> ?'

If you don't tell us what you want to do with it, what can we tell you.

Does one need to add two integers?

Excuse my ignorance, but why not just send the value as a string and
convert? No endianness to worry about and isnt that how the majority of
protocols operate, with string literals.

 

[Adie]

No. float types can work perfectly well without being converted,
just use the appropriate math libraries and output functions.

But then, if you were asking about networking stuff, which is
platform dependent, you will have to ask in a newsgroup that
discusses your platform. Networking stuff is not a part
of the _standard_ C++ language and thus not discussed here.

you'll be so fucking bugged out when byte ordering is added to the
standard.

 

[Dan Cernat]

If it does, and it is going between a little endian and a big endian
machines can we byte swap it as an int or long? Or does the mantissa or
exponent cross 2 byte boundarys?

Can I just call htonl ( (long) floatvalue))

The right question would be: 'does a float need to be converted to network
order if one want to <insert your action here> ?'

If you don't tell us what you want to do with it, what can we tell you.

Does one need to add two integers?


Dan

 

[JeffK.]

Look up the topic XDR / RPC -- its a C/C++ like language that addresses
data transfer between computers and eliminates the need to worry about
such
issues.

Hope this helps,

JeffK

 

[Adie]

In this specific case this will never happen.
Sending float's (or double's) in binary to a different
machine is asking for troubles. Byte ordering is the
least of your problems.

Ok Karl, I'll take your word for it.

 

[Alexander Terekhov]

Ok Karl, I'll take your word for it.

Take it with a pinch of salt. Little Karl never heard of IEEE 754,
I guess.

regards,
alexander.

 

[Alexander Terekhov]

Karl Heinz Buchegger wrote:
[...]

And yes: I have already been biten by trying to pass floating
point values in a binary form (PC to VAX, 12 years ago)

http://h18003.www1.hp.com/hps/ipf-enterprise/download/openvms_ipf_floating_point_wp.pdf

regards,
alexander.

 

[Steve Coleman]

Please show me in the C++ standard where it is mentioned
that every C++ implementation has to use that.

I haven't read the 'C++ standard' on this issue but I believe this is a
networking library question and thus the question should be asked in
another group.

But having to deal with this very issue on a daily basis I suggest that
they look into XDR or CDR as an answer to the origional problem. Any OS
worth is weight will have an implimentation of one or both of these (and
you can download/compile it if its not) and the libraries 'standard'
dictates that the library will perform the correct endian conversion on
both ends of the connection for all defined types, because thats what
its for!

Now it we wanted to discuss C++ wrappers for the network libraries...

;-)

 

[Karl Heinz Buchegger]

Both XDR and CDR use >>THE IEEE 754<< for representations.

Ahh. Now that's really usefull information. Not that I am
involved into network tasks. But those things could be
usefull some day. Thanks.

 

[Alexander Terekhov]

Ahh. Now that's really usefull information. Not that I am
involved into network tasks. But those things could be
usefull some day. Thanks.

Karl, only for you.

http://www.alphaworks.ibm.com/tech/decNumber

regards,
alexander.

P.S. And, BTW, it's not really expensive. ;-)